Home
About me

Kancheti Sai Srinivas

alt text 

Kancheti Sai Srinivas
Senior Researcher
Fujitsu Research India
Contact: Email, (+91) 9003107318, Google Scholar, LinkedIn

About

I was a final year Ph.D. student at IIT Hyderabad, advised by Prof. Vineeth N Balasubramanian, and have defended my thesis in the Department of Computer Science and Engineering at IIT Hyderabad (IITH). I have begun my new role as Senior Researcher at Fujitsu Research India. My research focuses on trustworthy machine learning, multimodal reasoning, and self-evolving agents. I am currently exploring memory for self-evolving agents. You can find my CV here.

Research Interests

  • Self-evolving agents and memory

  • Visual reasoning in Multimodal LLMs (spatial reasoning, grounding, RLVR/RL post-training)

  • Human-in-the-loop / preference-aware learning (algorithmic recourse)

  • Robustness, interpretability, and reliability of ML systems

News

Publications

Here is my Google Scholar

  • [2026] Faithful GRPO: Improving Visual Spatial Reasoning in Multimodal Language Models via Constrained Policy Optimization – Sai Srinivas Kancheti, Aditya S Kanade, Rohit Sinha, Vineeth N Balasubramanian, Tanuja Ganu. Conference on Language Modeling (COLM), 2026. Conference Paper

  • [2026] Chain-of-Thought Degrades Visual Spatial Reasoning Capabilities of Multimodal LLMs – Sai Srinivas Kancheti, Aditya Sanjiv Kanade, Vineeth N. Balasubramanian, Tanuja Ganu. Association for Computational Linguistics (ACL, Short Papers), 2026. ACL Anthology

  • [2026] Mind’s Eye: A Benchmark of Visual Abstraction, Transformation and Composition for Multimodal LLMs – Rohit Sinha, Aditya Sanjiv Kanade, Sai Srinivas Kancheti, Vineeth N. Balasubramanian, Tanuja Ganu. Association for Computational Linguistics (ACL, Long Papers), 2026. ACL Anthology

  • [2025] Efficient Vocabulary-Free Fine-Grained Visual Recognition in the Age of Multimodal LLMs – Hari Chandana K., K. Sai Srinivas, Gowtham Reddy A., Vineeth N. Balasubramanian. Transactions on Machine Learning Research (TMLR), 2025. OpenReview

  • [2025] HARE: Human-in-the-Loop Algorithmic Recourse – K. Sai Srinivas*, Rahul Vigneswaran*, Bamdev Mishra, Vineeth N. Balasubramanian. Transactions on Machine Learning Research (TMLR), 2025. OpenReview

  • [2025] Semantic Alignment for Prompt-Tuning in Vision Language Models – K. Sai Srinivas*, Hari Chandana K*, Gowtham Reddy A., Vineeth N. Balasubramanian. Transactions on Machine Learning Research (TMLR), 2025. OpenReview

  • [2025] Conformal Prediction in the Age of Multimodal Foundation Models: A Survey – K. Sai Srinivas*, Christian Stavan*, Kunal Tilaganji*, Sai Mathura Krishnan*, Vineeth N. Balasubramanian. Book chapter submission (2025).

  • [2024] Interpretable Model Drift Detection – Pranoy Panda, K. Sai Srinivas, V. Balasubramanian, Gaurav Sinha, CODS-COMAD ’24 link

  • [2022] Matching Learned Causal Effects of Neural Networks with Domain Priors – K. Sai Srinivas*, A. Gowtham Reddy*, V. Balasubramanian, Amit Sharma, ICML’22 link

Current Research

  • Exploring memory for self-evolving agents.

Education

  • Ph.D., Department of CSE, IIT Hyderabad (thesis defended; started Jan 2021; PMRF Scholar; GPA: 9.47)

  • Dual-Degree (BTech & MTech), Department of CSE, IIT Madras, Jul 2013 - Jul 2019 (Minor: Operations Research; GPA: 8.91)

Awards & Service

  • Microsoft India Research PhD Award (2024).

  • Prime Minister's Research Fellowship (PMRF) (2021).

  • Reviewer: TMLR; CVPR 2025 (top reviewer); AISTATS 2025; AAAI 2024/2025; NeurIPS 2024; ECCV 2024; subreviewer for ICML 2023/2024, IJCAI 2023, ICLR 2022/2023, AISTATS 2022.

  • Teaching Assistant: Deep Learning for Vision (Fall 2019, Spring 2021); Deep Learning (Spring 2022); Foundations of Machine Learning (Fall 2022); Advanced Topics in Machine Learning (Spring 2024).